fix: auto-correct model class for decoder models + show Colab logs in dashboard

fix: strip fences at PUSH_SCRIPT gate + stronger AI prompt for CausalLM

fix: strip markdown fences client-side + fix AI prompt for CausalLM models

fix: auto-detect decoder-only models, always use CausalLM for Qwen/GPT/Llama

fix: SQLite DB path → /tmp + channel ID fallback from Colab cache

fix: bridge uses async Supabase client — realtime requires it

feat: async job + polling architecture — fixes proxy timeout wall

fix: client-side template fallback + shorter timeouts — generation NEVER fails

fix: template fallback + 6 providers — script generation always works

fix: sync-first architecture — eliminates network errors

fix: restore max_tokens to 4096 + add retry back

fix: increase max_tokens to 1800 and strip markdown code fences

fix: eliminate retry-append duplication + partial-success detection

fix: stream recovery when connection drops mid-generation

fix: Qwen3-32B /no_think + skip reasoning_content tokens

fix: global SSE heartbeat + upgrade model to Qwen3-235B-A22B

fix: auto-retry + sync fallback for network errors in Architect

fix: multi-provider fallback, SSE heartbeat, timeout — kills network error

fix: embed bridge in notebook, pre-fill Supabase creds, add HTTP status ping

Fix: replace HfInference library with direct fetch to nscale provider (fixes fetch failed error)

Fix: detect env vars (Space secrets) in settings endpoint

🧠 Deploy NEXUS AI v1.0 — HF Inference powered